Block Is Building AI-Powered Security With OpenAI Technology, and It Shows Where Enterprise AI Is Heading Next

For the past few years, most companies have treated artificial intelligence as a productivity story. Write this email faster. Summarize this meeting. Draft this report. The wins were real, but they were mostly about saving minutes.

Block's move to build AI-powered security using OpenAI technology marks a different kind of bet. This is not AI as a helpful assistant sitting next to an employee. This is AI placed inside the systems that decide what is safe and what is not, the layer that watches for attacks, sorts signals from noise, and helps people respond before damage spreads.

That shift matters far more than it first appears. Security is one of the few areas where AI can be measured in avoided losses rather than saved keystrokes. And when a major fintech player builds this capability into its core operations, it sends a clear signal: AI is moving from the edges of the business into the machinery that protects it.

Why Security Is Becoming AI's Biggest Proving Ground

Security teams have a math problem that has nothing to do with talent and everything to do with volume. Modern organizations generate an enormous stream of alerts, logs, login attempts, device signals, and transaction patterns. Human analysts cannot read all of it. So they triage, and triage means something always gets skipped.

That is exactly the shape of problem large language models and AI systems are good at. They can read across huge amounts of messy, unstructured text, spot patterns that look out of place, and summarize what matters in plain language. A security analyst who once had to open a dozen dashboards can instead start with a written explanation of what is happening and why it might be a problem.

Block choosing to build this with OpenAI technology rather than with a narrow, single-purpose tool tells us something about how the company views the work. Security is not just pattern matching against a fixed list of known attacks. It is judgment, context, and communication, the ability to explain why something is suspicious to the person who has to make a call at 2 a.m.

The alert fatigue problem finally meets its match

Anyone who has worked in security operations knows the phrase "alert fatigue." It is the reason good analysts burn out. When a system cries wolf hundreds of times a day, people learn to stop listening.

AI changes that dynamic in two ways. First, it can group related signals so a hundred alerts become one story. Second, it can add context that a static rule never could, is this login unusual for this person, at this time, from this location, with this device history? That kind of layered reasoning is where generative AI has improved dramatically, and it is why security has become one of the fastest-moving areas of enterprise adoption.

From Watching to Acting: How the Job Changes

There is a subtle but important difference between AI that observes and AI that acts. Watching is safe and easy to sell. Acting is where the real value, and the real risk, lives.

Expect three stages to unfold across the industry in the coming year:

• Assist: AI reads, summarizes, and ranks. Humans still make every decision, but they make it faster and with better information.
• Recommend: AI suggests a specific response, block this account, revoke this token, isolate this device, and a human approves with one click.
• Automate: AI handles low-stakes, high-confidence responses on its own while escalating anything ambiguous.

Most mature programs will live in the middle stage for a while. That is not a weakness; it is the correct place to be. The value of a recommendation engine is enormous even when a human keeps the final say, because the human's time is freed for the cases that genuinely need a person.

The new security skill set

As AI takes over triage, the job of the security professional shifts toward verification, judgment, and system design. The people who thrive will not be the ones who can read the most logs fastest. They will be the ones who can ask a model the right question, spot when an answer sounds confident but is wrong, and design workflows where a mistake gets caught before it becomes an outage or a breach.

This is an underrated point for businesses planning headcount. The goal is not fewer security people. It is security people pointed at higher-value problems.

The Other Side of the Coin: Attackers Get the Same Tools

No honest discussion of AI in security can skip this. The same capabilities that help defenders are available to attackers, often at low cost and with fewer rules to follow.

Phishing messages that once contained obvious broken grammar now read naturally. Social engineering campaigns can be generated at scale and personalized. Reconnaissance that took days can be drafted in minutes. The volume and quality of attacks both go up.

That creates a strange competitive dynamic. Defense gets better, but the attack surface grows at the same time. This is why AI-powered security is not a project with an end date, it is an ongoing operating capability. The organizations that treat it as a one-time deployment will fall behind. The ones that treat it as a continuous loop, detect, learn, adjust, will compound their advantage.

Practical Implications for Businesses

For leaders watching this trend, the lessons are concrete and can be applied regardless of company size.

1. Start where the pain is measurable

Do not launch an "AI security program" in the abstract. Find the workflow that already hurts: the alert queue nobody clears, the manual review that takes four days, the access requests that pile up. Measure the baseline. AI projects succeed when the before-and-after is visible in a number.

2. Keep humans in the loop, but design the loop on purpose

"Human in the loop" is often said but rarely designed. Ask the hard questions up front: What exactly does the human review? How much context do they get? What is the escalation path when the model is unsure? A loop that nobody actually checks is not a safeguard; it is a formality.

3. Treat data handling as a security decision

Feeding internal logs and incident data into an AI system means data leaves its original boundaries. Before deploying, be clear about what is sent, what is retained, what is used for training, and who can access the results. For regulated industries, this conversation should involve legal and compliance from day one, not after the pilot.

4. Plan for the model to be wrong

AI systems will produce false positives and false negatives. That is not a bug to be eliminated; it is a property to be managed. Build feedback channels so analysts can flag bad outputs, and use that signal to tune prompts, rules, and thresholds. The organizations that get good at this loop pull ahead quickly.

5. Defend the AI itself

An AI system that reads untrusted input is a new attack surface. Instructions hidden in a document, a log entry, or a message can try to steer the model. Security teams should treat every model with the same suspicion they treat any other internet-facing service, test it, bound it, and limit what it is allowed to do.

6. Budget for the boring parts

Logging, audit trails, access control, and documentation rarely make headlines, but they are what allow a security AI deployment to survive a review, an audit, or an incident. The flashy demo is the easy part.

What This Means for the Future of AI

Block's investment points toward a broader pattern that will define the next phase of enterprise AI.

First, AI is moving toward consequential work. The first wave was low-risk and high-visibility, chat, drafting, summarizing. The next wave is high-stakes and mostly invisible: fraud detection, identity verification, incident response, compliance monitoring. The bar for reliability, explainability, and auditability is far higher, and vendors and builders will be judged on those terms.

Second, security may become the clearest way to prove AI's return on investment. Unlike many AI use cases, security has a natural scoreboard: losses prevented, incidents contained, time to resolution. That makes it easier for technology leaders to defend budgets and easier for regulators and boards to see value.

Third, the partnership model matters. Building on a major model provider's technology rather than training from scratch is a pragmatic choice that many organizations will copy. It trades some control for speed. The companies that do it well will be the ones that invest heavily in the layer around the model, the policies, the evaluation, the guardrails, the human workflow, because that is where differentiation actually lives.

Fourth, trust becomes a product feature. As AI takes on a larger role in protecting customers, the ability to explain what the system did and why will matter as much as the outcome. A customer locked out of an account by an automated decision needs a fast, clear path to resolution. That is as much a design problem as an engineering one.

The Road Ahead

Security has always been a game of adaptation. Every new defensive tool prompts attackers to find the edges, and every new attack pushes defenders to build something better. AI accelerates both sides of that loop.

What makes Block's direction notable is not any single capability. It is the recognition that AI is not a side experiment to be bolted onto a security team, it is becoming part of the security team. Companies that treat it that way, with clear guardrails, honest measurement, and a disciplined human workflow, will find themselves better protected and better prepared for whatever comes next.

The organizations that wait for the technology to feel completely safe before adopting it will be making a bet of their own: that their attackers are also waiting. That bet rarely pays off.